HAL Id: hal-01540956
https://hal.inria.fr/hal-01540956v3
Submitted on 5 Dec 2019
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
space-time domain decomposition for two-phase flow between different rock types
Elyes Ahmed, Sarah Ali Hassan, Caroline Japhet, Michel Kern, Martin Vohralík
To cite this version:
Elyes Ahmed, Sarah Ali Hassan, Caroline Japhet, Michel Kern, Martin Vohralík. A posteriori error
estimates and stopping criteria for space-time domain decomposition for two-phase flow between differ-
ent rock types. SMAI Journal of Computational Mathematics, Société de Mathématiques Appliquées
et Industrielles (SMAI), 2019, 5, pp.195-227. �10.5802/smai-jcm.47�. �hal-01540956v3�
for space-time domain decomposition
for two-phase flow between different rock types ∗
Elyes Ahmed †‡§ Sarah Ali Hassan † Caroline Japhet ‡ Michel Kern † Martin Vohral´ık †
Abstract
We consider two-phase flow in a porous medium composed of two different rock types, so that the capillary pressure field is discontinuous at the interface between the rocks. This is a nonlinear and degenerate parabolic problem with nonlinear and discontinuous transmission conditions on the interface.
We first describe a space-time domain decomposition method based on the optimized Schwarz waveform relaxation algorithm (OSWR) with Robin or Ventcell transmission conditions. Complete numerical approximation is achieved by a finite volume scheme in space and the backward Euler scheme in time.
We then derive a guaranteed and fully computable a posteriori error estimate that in particular takes into account the domain decomposition error. Precisely, at each iteration of the OSWR algorithm and at each linearization step, the estimate delivers a guaranteed upper bound on the error between the exact and the approximate solution. Furthermore, to make the algorithm efficient, the different error components given by the spatial discretization, the temporal discretization, the linearization, and the domain decomposition are distinguished. These ingredients are then used to design a stopping criterion for the OSWR algorithm as well as for the linearization iterations, which together lead to important computational savings. Numerical experiments illustrate the efficiency of our estimates and the performance of the OSWR algorithm with adaptive stopping criteria on a model problem in three space dimensions. Additionally, the results show how a posteriori error estimates can help determine the free Robin or Ventcell parameters.
Key words: two-phase Darcy flow, discontinuous capillary pressure, finite volume scheme, domain de- composition method, optimized Schwarz waveform relaxation, Robin and Ventcell transmission conditions, linearization, a posteriori error estimate, stopping criteria
1 Introduction
Two-phase flows in porous media are of interest in many applications, such as CO2 sequestration in saline aquifers, description of oil reservoirs, or gas migration around a nuclear waste repository in the subsurface.
The numerical simulation of such flows is a challenging task. One well-known reason, which is the main topic of this paper, is that the capillary pressure and the relative permeability functions may be discontinuous across the interface between different regions of the domain.
We consider in this paper a simplified two-phase flow model (one equation, no advection) introduced in [24] to study the phenomenon of oil or gas trapping in a porous medium with several rock types (see
∗ This work was supported by the French ANR DEDALES under grant ANR-14-CE23-0005, Labex MME-DII, and the French Agency for Nuclear Waste Management (ANDRA). It also received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (grant agreement No 647134 GATIPOR).
† Inria, 2 rue Simone Iff, 75589 Paris, France & Universit´ e Paris-Est, CERMICS (ENPC), 77455 Marne-la-Vall´ ee, France.
michel.kern@inria.fr, martin.vohralik@inria.fr
‡ Universit´ e Paris 13, Sorbonne Paris Cit´ e, LAGA, CNRS UMR 7539, 93430, Villetaneuse, France.
japhet@math.univ-paris13.fr
§ Department of Mathematics, University of Bergen, Bergen, Norway. elyes.ahmed@uib.no
§ UPMC Sorbonne Universit´ e, Laboratoire Jacques-Louis Lions, B.C. 187, Paris, F-75005 France.
sarah.alihassan.upmc@gmail.com
1
also [14, 52] for the analysis of a 1D model). The main theoretical and numerical difficulties of this problem are the nonlinearity and degeneracy of the parabolic equation for the saturation of one of the phases, together with nonlinear and discontinuous transmission conditions at the interface between the rocks. Existence of a weak solution has been proven in [24], using the convergence of a finite volume scheme (see also [17, 18], and [15] for the full two-phase flow model). Uniqueness of a weak solution has been proven in [17] for a particular choice of functions characterizing the porous medium and in [18, 19] for the full two-phase problem with advection terms in the one-dimensional case. An a priori convergence study of a finite volume discretization has been undertaken in [24]. Due to different hydrogeological properties of the different rocks, domain decomposition (DD) methods appear to be a natural way to solve efficiently two-phase flow problems, and are proposed in [55, 56] as well as [31, 39, 48, 49, 50, 51].
The first purpose of this paper is to propose a domain decomposition method that is global-in-time, with time-dependent, nonlinear, and discontinuous optimized Ventcell transmission conditions at the interface between rock types, for solving the two-phase flow model of [24]. This method is based on the optimized Schwarz waveform relaxation algorithm (OSWR), in which at each OSWR iteration, space-time nonlinear subdomain problems over the whole time interval are solved. The exchange between the subdomains is here ensured using the nonlinear Ventcell transmission operators. Crucially, this approach allows for different time stepping in different parts of the domain, adapted to the physical properties of each subdomain, and for parallelization in time, in contrast to the domain decomposition algorithms discussed above. We are not aware of any domain decomposition algorithm proposed and analyzed for simultaneously a) degen- erate parabolic problems; b) nonlinear and discontinuous transmission conditions; c) Robin and Ventcell transmission conditions; d) global-in-time formulation. Some of these ingredients have appeared previously independently as in [12, 13, 15, 32, 33, 35] and the references therein. The use of higher-order (Ventcell) transmission operators allows physically more valuable information to be exchanged between the subdo- mains and hence typically leads to a better convergence behavior, see [12, 29, 33, 35, 36, 37, 41] and the references therein for linear problems, and [16, 32] for problems with nonlinear reaction terms. The Robin case, obtained by setting one of the Ventcell parameters to zero, is analyzed in [3, 4], both for the exis- tence of a weak solution of the subdomain problem with Robin boundary conditions, and to prove that the space-time DD method is well-defined.
The second purpose of this paper is to derive a posteriori error estimates for the finite volume – backward Euler approximation of the proposed space-time DD algorithm. We build here on the (few) previous contributions on unsteady, nonlinear, and degenerate problems in [20, 23, 42]; we will in particular rely on [23, Theorem 5.2], where the degenerate nature of the problem is handled to obtain an energy-type upper bound on the error. Two additional specific difficulties that were to the best of our knowledge not treated previously are that the current problem features nonlinear and discontinuous transmission conditions and that our finite volume discretization leads to an approximate solution that is nonconforming in space.
Finally, the third purpose of this paper is to address the question of when to stop the domain decom- position iterations, as well as the iterations of the linearization solver used for the subdomain problems. In contrast to prevalent practice, where one iterates until some fixed tolerance has been reached, we propose to stop when the DD/linearization error component is up to a user-specified fraction below the total error.
This typically spares numerous iterations. In practice, this means that the a posteriori error estimates we develop have to hold true at each iteration of the OSWR algorithm and the linearization, and distinguish the different error components (space, time, linearization, and domain decomposition). For this part of our work, we take up the path initiated in [9, 11, 26, 38, 54] for general techniques taking into account inexact algebraic solvers, [45] for (multiscale) mortar techniques, [46, 47] for FETI and BDD domain decomposition algorithms combined with conforming finite element discretizations, and most closely [6, 7] for respectively steady and unsteady linear problems with similar space and time discretizations. To achieve our goals here, we use a) equilibrated H (div)-conforming flux reconstructions, piecewise constant in time, that are direct extension of [7, 25, 54] to our model; b) novel subdomain-wise H 1 -conforming saturation reconstructions, continuous and piecewise affine in time, which are targeted to the present setting in that they satisfy the nonlinear and discontinuous interface conditions.
The outline of the paper is as follows: Section 2 recalls the physical model and defines weak solutions as
well as the relevant functions spaces. Section 3 presents the space-time domain decomposition method with
nonlinear and discontinuous Ventcell transmission conditions. In Section 4, we present the discrete OSWR
algorithm, by combining a finite volume scheme for the discretization in the individual subdomains and the
backward Euler time stepping with the OSWR method. We then construct the needed ingredients for the
a posteriori error estimates: Section 5 defines the postprocessing as well as the H 1 - and H (div)-conforming reconstructions. Section 6 puts the pieces together by presenting a guaranteed and fully computable error estimate that bounds the error between the unknown exact solution and the approximate solution. In Section 7 we decompose this overall estimator into individual estimators characterizing the space, time, domain decomposition, and linearization error components. This is used in Section 8 to propose stopping criteria for the OSWR algorithm and for the nonlinear iterations. The method is numerically validated on three examples in three space dimensions in Section 9. Finally, Section 10 draws some conclusions.
2 Presentation of the problem
Let Ω be an open bounded domain of R d , d = 2 or 3, which is assumed to be polygonal if d = 2 and polyhedral if d = 3. We denote by ∂Ω its boundary (supposed to be Lipschitz-continuous) and by n the unit normal to ∂Ω, outward to Ω. Let a time interval (0, T ) be given with T > 0. We consider a simplified model of a two-phase flow through a heterogeneous porous medium, in which the advection is neglected.
Assuming that there are only two phases occupying the porous medium Ω, say gas and water, and that each phase is composed of a single component, the mathematical form of this problem as it is presented in [17, 24] is as follows: given initial and boundary gas saturations u 0 and g, as well as a source term f , find u : Ω × [0, T ] → [0, 1] such that
∂ t u − ∇· (λ(u, x) ∇ π(u, x)) = f, in Ω × (0, T ), (2.1a)
u( · , 0) = u 0 , in Ω, (2.1b)
u = g, on ∂Ω × (0, T ). (2.1c)
Here u is the gas saturation (and therefore (1 − u) is the water saturation), π(u, x) : [0, 1] × Ω → R is the capillary pressure , and λ(u, x ) : [0, 1] × Ω → R is the global mobility of the gas given by
λ(u) = λ w (1 − u)λ g (u) λ w (1 − u) + λ g (u) ,
where λ w and λ g are the phase mobilities. One can refer to [20, 21, 24] for a derivation of (2.1a)– (2.1c) from the complete two-phase flow model. For simplicity, we consider only Dirichlet boundary conditions on
∂Ω. Other types of boundary conditions could be dealt with the same way as in [17, 24, 54]. The model problem given by (2.1a) is a nonlinear degenerate parabolic problem as the global mobility λ(u) → 0 for u → 0 and 1, and, moreover, π 0 (u) → 0 for u → 0 (see [10, 21]).
2.1 Flow between two rock types
In this part, we particularize the model problem (2.1a) to a porous medium with different capillary pressure curves π i in each subdomain, following [24]. We suppose that Ω is composed of two disjoint subdomains Ω i , i = 1, 2, which are both open polygonal subsets of R d with Lipschitz-continuous boundary. We denote by Γ the interface between Ω 1 and Ω 2 , i.e., Γ = (∂Ω 1 ∩ ∂Ω 2 ) ◦ . Let Γ D i = ∂Ω i ∩ ∂Ω. Both data λ and π, which can in general depend on the physical characteristics of the rock, are henceforth supposed to be homogeneous in each subdomain Ω i , i = 1, 2, i.e., λ i ( · ) := λ |Ω i ( · ) = λ( · , x), ∀ x ∈ Ω i , and similarly for π i . Equations (2.1a) in each subdomain Ω i then read as
∂ t u i − ∇· (λ i (u i ) ∇ π i (u i )) = f i , in Ω i × (0, T ), (2.2a)
u i ( · , 0) = u 0 , in Ω i , (2.2b)
u i = g i , on Γ D i × (0, T ). (2.2c) We use the notation v i = v | Ω i for an arbitrary function v.
Before transcribing the transmission conditions on the interface Γ, we make precise the assumptions on the data (further generalizations are possible, bringing more technicalities):
Assumption 2.1 (Data). 1. For i ∈ { 1, 2 } , π i ∈ C 1 ([0, 1], R) can be extended in a continuous way to a function (still denoted by π i ) such that π i (u) = π i (0) for all u ≤ 0 and π i (u) = π i (1) for all u ≥ 1.
Moreover, π i | [0,1] is a strictly increasing function.
2. For i ∈ { 1, 2 } , λ i ∈ C 0 ([0, 1], R + ) is bounded and can be extended in a continuous way to a function (still denoted by λ i ) such that λ i (u) = λ i (0) for all u ≤ 0 and λ i (u) = λ i (1) for all u ≥ 1 . We denote by C λ an upper bound of λ i (u), u ∈ R .
3. The initial condition is such that u 0 ∈ L ∞ (Ω) with 0 ≤ u 0 ≤ 1 a.e. in Ω.
4. The boundary conditions 0 ≤ g i ≤ 1 are traces of some functions in L 2 (0, T ; H 1 (Ω i )). For simplicity, we suppose them to be at most piecewise second-order polynomials with respect to the boundary faces of the spatial mesh introduced in Section 4.1.1 below, continuous on Γ D i , and constant in time. Moreover, they need to match in the sense that π 1 (g 1 (x)) = π 2 (g 2 (x)) for all x ∈ Γ ∩ Γ D 1 ∩ Γ D 2 .
5. The source term is such that f ∈ L 2 (0, T ; L 2 (Ω)). For simplicity we further assume that f is piecewise constant in time with respect to the temporal mesh introduced in Section 4.1.2 below.
For simplicity, we suppose in Assumption 2.1(4) that the boundary conditions are piecewise polynomial in space and constant in time, so that there is in particular no additional data oscillation error stemming therefrom.
We give now the transmission conditions needed to connect the subdomain problems (2.2), for i = 1, 2.
We consider two cases. The first case is when
π 1 (0) = π 2 (0) and π 1 (1) = π 2 (1), (2.3)
the same way as in [17]. If the functions π i satisfy the above condition, the capillarity curves are said to be matching and the resulting transmission conditions on the interface are given by
π 1 (u 1 ) = π 2 (u 2 ), on Γ × (0, T ), (2.4a) λ 1 (u 1 ) ∇ π 1 (u 1 ) · n 1 = − λ 2 (u 2 ) ∇ π 2 (u 2 ) · n 2 , on Γ × (0, T ). (2.4b) These conditions yield a discontinuous saturation across the interface, i.e., we find that in general u 1 6 = u 2
on Γ.
In the second case, i.e., in the case when
π 1 (0) 6 = π 2 (0) or π 1 (1) 6 = π 2 (1), (2.5) the capillarity pressure curves are said to be non-matching. Consequently, not only the saturation is discontinuous at the medium interface, but also the capillary pressure field. The condition (2.5), studied in [24], has direct consequences on the behavior of the capillary pressures on both sides of the interface Γ. Indeed, suppose that π 1 (0) ≤ π 2 (0) < π 1 (1) ≤ π 2 (1), that u ∗ 1 is the unique real in [0, 1] satisfying π 1 (u ∗ 1 ) = π 2 (0), and that u ∗ 2 is the unique real in [0, 1] satisfying π 2 (u ∗ 2 ) = π 1 (1). Then, if u 1 ≥ u ∗ 1 and u 2 ≤ u ∗ 2 , we can still prescribe the connection of the capillary pressures on the interface Γ π 1 (u 1 ) = π 2 (u 2 ) as in (2.4a). If 0 ≤ u 1 ≤ u ∗ 1 , the model imposes u 2 = 0, and the gas phase is entrapped in the rock Ω 1 while the water flows across Γ. In the same way, if u ∗ 2 ≤ u 2 ≤ 1, the model prescribes u 1 = 1, and the water phase is captured in Ω 2 as a discontinuous phase while the gas flows across Γ (see Fig. 1 left). Following [21, 24], these conditions on the gas-water saturations on the interface Γ are simply given by
π 1 (u 1 ) = π 2 (u 2 ), on Γ × (0, T ), (2.6a) λ 1 (u 1 ) ∇ π 1 (u 1 ) · n 1 = − λ 2 (u 2 ) ∇ π 2 (u 2 ) · n 2 , on Γ × (0, T ), (2.6b) where π i , for i = 1, 2, are truncated capillary pressure functions given on [0, 1] respectively by π 1 : u 7→
max(π 1 (u), π 2 (0)) and π 2 : u 7→ min(π 2 (u), π 1 (1)) (see Fig. 1 right). In [24], it has been established that the model problem (2.2) together with the transmission conditions (2.6) has the necessary mathematical properties to explain the phenomena of gas trapping (see also [14, 8, 19, 52]).
2.2 Transformation of the equations and weak formulation
Still following [24], we present here the mathematical quantities and function spaces used to characterize
the weak solution to the multidomain problem (2.2) with the conditions (2.6). That of the problem (2.2)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0
1 2 3 4 5 6 7
u
1∗ π
1( u )
π
1(0) π
1(1)
u
1π
2(0)
π
2( u ) π
2(1)
u
2u
∗20 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0 1 2 3 4 5 6 7